Mailing List Archives
Authenticated access
|
|
|
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Condor-users] Dynamic provisioning prevents Evenly Divided Resources tasks?
- Date: Tue, 2 Aug 2011 16:17:32 +0200
- From: Lukas Slebodnik <slebodnik@xxxxxxxx>
- Subject: Re: [Condor-users] Dynamic provisioning prevents Evenly Divided Resources tasks?
Hi,
If you don't write request_* to your submit file than Condor will use default values.
default for request_cpu is 1,
default for request_disk is the attribute DiskUsage,
default for request_memory is based on ImageSize or JobVMMemory.
request_disk is done in Kbytes. If dynamic provisioning is enabled for the
slot, and this submit command is not specified, then for non-vm-universe jobs,
it defaults (for the initial value of the job attribute DiskUsage) to the size
of the executable plus the sum of input file sizes. For vm-universe, it
defaults to the size of the disk image.
request_memory is in Mbytes. It sets an intial value for ImageSize if specified
(for non-vm-universe jobs). If not specified, initial value of ImageSize is the
size of the executable. For vm universe jobs, ImageSize is set by the submit
cmd vm_memory, and if not specified ImageSize is initialized to 0. For the case
of a vm universe job that (incorrectly) has both request_memory and vm_memory
set, the value given by vm_memory is used.
But you can automatically apend expresions to the job ClassAds using
SUBMIT_EXPRS settings to override default values.
For example:
#default value for request_memory will be 2048 MB
RequestMemory = 2048
SUBMIT_EXPRS = $(SUBMIT_EXPRS) RequestMemory
Regards,
Lukas
On Tue, Aug 02, 2011 at 09:00:11AM -0500, Edier Zapata wrote:
> Good morning,
> I have a question about Dynamic proviosing:
> If I configure my nodes to use it, and I want to run jobs in the "normal
> way" I mean, each job will use 1 core and 1/NumberOfCores resources I always
> have to add:
> request_cpus =1
> request_memory=XXXX
> request_disk=XXXX
> to the submit file?
> Or I can have both configurations in the same node?
>
> Thank you.
> --
> Edier Alberto Zapata Hernández
> Ingeniero de Sistemas
> Universidad de Valle